The automation paradox | Making training relevant

I stole the title, but the picture is all mine. Let's talk a little about what is going on in that picture: Right there is a very fine pilot. She is methodical, skilled, excellent at influencing her colleagues when needed, great at fitting into the team just about right (the trickiest of tasks for the second in command). Just left of centre you can see a white, plastic lever. This lever is poking up. It is surrounded by two others. These are the three autopilot engage/disengage levers. The 'Center' autopilot is engaged… or at least that's what it looks like from here.

That's one of my favourite pictures to date. It reminds me what a great view we had during that approach: A nice flight down from Edinburgh. Great weather, no aircraft defects or complication. We had briefed the approach and I offered my FO the option of a manually flown approach (remember, this is BA, so the landing is mine but the approach is flown by the other pilot). We discussed some of the threats and how we might mitigate them.

“Okay, so what if the workload ramps up and we decide that a manually flown approach isn't such a great idea?”

Seems like an easy fix? Stick the autopilot back in, right? Let the automatics take the strain. It isn't a bad start but what is that really going to look like if it happens? How are we going to make that all work? If I told you that more than one crew has ended up in precisely the same scenario with only half of the automatic systems controlling the aircraft would you believe me?

The 767 isn't like the 777 or the 787. When the autopilot is disengaged, we disconnect the autothrottle too. The autothrottle just isn't quite as clever as newer jets which are designed to help out when pitch and roll are controlled manually. The 767 is from a generation back. Manual flying during an approach means a fistful of thrust levers, too. On takeoff, however, the thrust is set with the touch of a button. The autothrottle system then winds the levers up for you. The nature of the early part of every flight means that you will be flying with a more-or-less fixed thrust setting right up until you reach up and push that middle lever up to engage the autopilot.

That's where the first threat originates: for your whole conversion course and for hundreds of flights the muscle memory for 'automatics in' equates to one switch selection. If your starting point isn't a takeoff then two selections are needed. One to get the autopilot working and another to tell the autothrottle what to do. So it isn't that surprising that when the workload is high we become hostages to our base instincts and revert to what feels normal… that old, familiar, single selection. No autothrottle means no thrust, which means no speed control for the manoeuvre you are trying to fly.

But Boeing cannot have created such a precarious device, can it? One where a single omission can point the way to disaster? Correct… they haven't: Switches and levers are just part of the equation. For decades now, the airline pilot has been in control of a highly integrated machine with sophisticated feedback methods. The modern airliner doesn't just listen and obey… it talks back.

Spread right across the middle of the flight deck is a collection of buttons and selectors which control the automation – it's known as the Mode Control Panel (or MCP). Airbus call it something different but the principle is identical. This part of the aircraft is where the pilot makes most automation selections; it's how we talk to the aircraft. The aircraft talks back through a line of electronic icons and captions on the main flying instrument. In other words, aircraft behaviour is continuously presented to both pilots. This part of the instrumentation is known as the Flight Mode Annunciator (FMA).

Forgetting to re-engage the autothrottle system would be obvious on the FMA long before the aircraft ended up in danger and perhaps stalling or flying slowly into the ground; a big, fat hole in the neat arrangement of mode captions. Our brief discussed becoming prey to our human failings. The countermeasure was to remind ourselves to stay on top of FMAs; to rehearse physical selections we would make and what we were expecting to happen on the displays.

I have stolen the title of this piece from an article in the Economist. The article was pretty good. Quite well researched but it drew some odd conclusions. I could emphathise with them because they are formed in the wake of accidents which seem to defy common sense. Asiana, UPS both seem to point towards 'overly complex' automation causing problems for crews. The hapless crews seemingly become bewildered by endless modes, functions and submodes. Perhaps it would be better to get rid of all of this cleverness and go back to basics?

The first assumption in this 'Automation paradox' seems to be that more technology should mean better, safer airliners. We shouldn't kid ourselves. By and large, aviation is far safer because of automated, integrated technologies including autopilots. To really understand what is going on here we need to go for a ride through the last generation of airliners, training and culture to pin the problem down. Is it even a paradox?

Why don't we start with the VC10… Spiritual predecessor to my own 767 at BA. There are some great pictures of the mighty VC10 at Entebbe. Every time I fly there, I am sure I can see the ghost of these graceful old beasts clawing into the air over Lake Victoria. I can almost imagine the thunderous crackle of the four engines burning long into the soggy African sky.

I had the opportunity to visit Brize Norton, get up close to a real VC10 and fly the VC10 simulator. In contrast to modern airliners automation was secondary to the business of flying the beast and managing its many systems. I have included a picture of my limp and sweaty body hunched over the controls flying a teardrop to the ILS. The automation is arranged behind the thrust levers. Nowhere near my normal field of view. Monitoring the automation was purely a matter of monitoring aircraft performance. Very little mode information was available. Autothrottle was actually behind you. A whole human being was tasked with setting thrust or chasing airspeed. A little strange at first but quite nice in the end. Dan, my colleague suggested that it was “…rather like flying with a butler…”.

So, this is our starting point. A 50 year old aircraft representing the very beginning of high-performance, automated jet transports. 4 aviators flying the aircraft (I forgot to mention the navigator before) using simple automation methods in a low-denisity, beacon-to-beacon aviation environment. At the time basic skills standardisation was elusive and a number of high profile accidents forced the regulator to take action to ensure that all pilots (even the co-pilot!) could fly a simple selection of common manouevres and approaches.

The complicated VC10 was a technical monster. Valves, pumps, ducts, crossfeeds, acronyms, breakouts, dolls' eyes. Quite apart from actually flying the thing, you had to know it all. So groundschool was designed to teach you. In great detail, with great accuracy. You needed to know all of this stuff to overcome design shortcomings like gauges with no limit markings; strange interactions between systems; complex ergonomics making it difficult to spot omissions.

Frequently, groundschool instructors were engineers. Their job was to understand the aircraft inside out and to get you through an exam written by a similarly technocratic individual. Groundschool tended to steer well clear of anything considered 'pilot territory'. Groundschool has not changed much.

Here is the first industry problem: Find a paper copy of the technical manual for the 767 (a 1982 aircraft). Hold it so that you are looking at the subject dividers and then search for the thickest section. When you look closer you will find that the 'Autoflight' and 'Flight Management and Navigation' section are the biggest parts. That tells you something about the aircraft: Automation is really important.

I recall doing my first EFIS jet course, the 737-400, long before I worked for BA. The groundschool instructor had no practical idea about mode logic or autopilot functions, and consequently, neither did we. We learned some mode-related information by rote with one objecive in mind: Pass the exam. Knowing almost nothing about the way in which the automation worked, I graduated to the sim phase. It turned out that a lot of the functions were pretty intuitive although the VNAV modes were a bit of a mystery. The simulator was a blur of learning SOPs, coping with the speed and handling of a modern jet transport aircraft. The regulatory target was very clear: pass the licensing check. This was – and still is – a handling exercise. Manual skill is under the microscope. And fair enough – we all want our pilots to actually be able to fly the aircraft.

It is more than 15 years since I stood outside Aviation House at Gatwick staring at the printed letters '737' in disbelief that I had managed to get a jet type on my licence. I suspect that VC10 pilots had undergone a similar experience years before me. The system had not really changed since back then. I have blogged before about 'trainee needs' [Staines Trident Lessons] and the truth is that I was well prepared to fly the 737 in a fairly basic fashion. Of course, I was destined to be a First Officer and expectations about my capability would be different to, say, a converting captain. I reckon that I had sufficient knowledge to put my captain in an uncomfortable situation and about 10% of what was required to help him out of it.

So, there's no paradox here. The traditional regulatory environment is a very limited tool rooted in history. It is designed around aircraft which were flown in different ways; like the VC10. Automation was not her biggest challenge. The VC10 demanded skill, cunning and guile to keep its many systems happily on the boil. If you are curious, it was a fairly easy aircraft to hand-fly. Heavy in roll when making big demands, but really quite light when making fine corrections. My capacity was sapped by the tiny, old fashioned flight instrumentation and completely random scattering of useful information around the flight deck.

I will blog a little more about automation handling soon because it is useful to understand how pilots must work through stages of automation that are appropriate to the job at hand. The reflexive response to Air France 447 and other 'loss of control' events was to cry out that pilots needed more time recovering from unusual attitudes. I am worried that the automation debate will move in the same way: take the autopilot out and fly manually. Boeing and IAFTP published a fantastic paper named 'Airline Pilot Perceptions of Training Effectiveness' right here. If you are reading this and you are in charge of any pilot training you really need to read that paper…

Here is a quote from the exec summary:

“…We conclude training could be improved to prepare pilots for their actual work by delivering content that is relevant to daily flight operations…”

The industry needs to address relevance in training right at the outset. I'm obviously going to blow my own trumpet because I reckon that at BA we are really good at making training relevant to the trainee and to the environment that the trainee will operate in.

Back to that picture right at the top of the article… a very fine pilot, flying in great weather – with the autopilot engaged. Why? Because it was appropriate to do so. Heathrow were using improving weather to pack lots of aircraft in. We used our professional judgement and decided that, on balance, our capacity would be best preserved to monitor the traffic situation.



Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s